[PyPANDA] BlockingIOError: [Errno 11] write could not complete without blocking

[Rubonnek]

Trying to dump the CPU registers for each instruction may cause a BlockingIOError:

@panda.cb_insn_translate
def should_run_on_insn(CPUState_env, target_ptr_t_pc):
    return True

@panda.cb_insn_exec
def on_insn(CPUState_env, target_ptr_t_pc):
    print("CPU STATE: ", CPUState_env)
    print(panda.arch.dump_regs(CPUState_env))
    return 0

Traceback (most recent call last):
  File "./pypanda_test.py", line 118, in <module>
    panda.run_replay("preinit_launch")
  File "/usr/local/lib/python3.8/dist-packages/pandare/panda.py", line 626, in run_replay
    self.run()
  File "/usr/local/lib/python3.8/dist-packages/pandare/panda.py", line 540, in run
    raise saved_exception
  File "/usr/local/lib/python3.8/dist-packages/pandare/panda.py", line 2581, in _run_and_catch
    r = fun(*args, **kwargs)
  File "./pypanda_test.py", line 114, in on_insn
    print(panda.arch.dump_regs(CPUState_env))
  File "/usr/local/lib/python3.8/dist-packages/pandare/arch.py", line 237, in dump_regs
    telescope(self.panda, cpu, val)
  File "/usr/local/lib/python3.8/dist-packages/pandare/utils.py", line 60, in telescope
    print()
BlockingIOError: [Errno 11] write could not complete without blocking

This seems to be an issue on non-blocking writes to stdout, which is set somewhere in the code.

[lacraig2]

I haven't been able to replicate your issue, but I will suggest that you change:

print(panda.arch.dump_regs(CPUState_env)

to:

panda.arch.dump_regs(CPUState_env)

as dump_regs calls print itself and does not return a value.

It may well be that the telescope call, which calls print, inside an existing print evaluation might run into a blocking IO issue? Hard to say.

[Rubonnek]

panda.arch.dump_regs(CPUState_env)

The issue still happens with this change.

[lacraig2]

Odd. Well it looks like stdout is blocking. You could set stdout to not block like in this issue:

import fcntl
fcntl.fcntl(1, fcntl.F_SETFL, 0)

Let me know if that makes the error go away. If so we can look at handling it at some other level.